0542-B1

African deforestation: causes and scenarios

Matti Palo and Erkki Lehto ¹

Abstract

Africa has about 528 million ha of various kinds of tropical natural forests. With 16 independent variables and relative forest area as a dependent variable we made regression modelling using 277 original year observations from FAO's FORIS database. They represent subnational geographical areas of 35 countries. In this modelling we controlled the variation of ecological conditions among subnational units by five ecological variables. It was also possible to control the variation of data reliability with three reliability variables. The two control measures we have not seen applied simultaneously by other deforestation modellers. Consequently, we are confident that our regression coefficients for population and income variables are less biased than without these two controls. The distribution of the relative forest areas is close to normal and the distribution of the residuals from our model estimation is quite random. Both of these qualities support the quality of our modelling. As a result we received empirical support for our hypothesis that increasing human population and income are respectively decreasing relative forest area at the low income levels prevailing in Africa. We made trend scenarios and seven alternative scenarios of tropical forest area decrease in Africa by our regression model with alternative assumptions of population and income growth. From 1990 to 2025 our scenarios indicate a reduction in forest area from one-third to a half. If true, this scale of deforestation would invalidate the use of natural forests in order to reduce the number of poor by 50% by 2015 - the Millennium objective of the United Nations. We assumed a transition in natural forest area decline to take place after 2025.

1. Introduction

In Africa forests occupy an area of 528 million hectares, most of it, 505 million hectares, are located in the 40 countries of tropical Africa. Forest cover varies greatly by countries and consists as an average of 23 percent of land area in tropical Africa in 1995. Forest is defined as an ecosystem with a minimum of ten percent of crown cover of trees and/or bamboos, generally associated as wild flora, fauna and natural soil conditions, and not subject to agricultural practices (FAO 1993). This broad definition decreases the scale of deforestation in comparison with, for example, if a crown cover of 40 percent had been applied (closed forests). The mix of forests in tropical Africa is composed of moist and dry deciduous forests, rainforests, very dry forests as well as of hill and montane forests.

African forests have been under anthropogenic influences since at least four millennia. Forests were cleared for cultivation and pasture in a wider scale via indigenous discoveries in iron making well before western influences. Large-scale deforestation took place in Africa well before the arrival of European colonialism (Siiriäinen 1996).

We define deforestation of natural forests as a clearing of forests of the above definition to other land use categories, primarily to agriculture, pasture, shrubs, plantations or infrastructure. In addition forest degradation is occurring. Degradation usually lowers the biomass and biodiversity. Our analysis excludes forest degradation and changes in plantation forests but concentrates only on deforestation of natural forests.

Tropical deforestation In Africa has been estimated by FAO (1993, 2001) as four to five million hectares as an annual average during the 1980's and 1990's. The countries with largest forests are deforesting most. In the similar way most deforestation is taking place in largest forest formations, such as moist and dry deciduous forests and rainforests.

Small-scale subsistence farming is the dominating local agent of African deforestation changing closed forests to short fallow or first closed forests to open forests and then to various agricultural uses. Fragmentation is also a major deforestation process driven by various local agents. We exclude local deforestation agents from our modeling but concentrate on the underlying causes, which we believe are most important in scenario making and as policy instruments.

A conservation atlas of tropical forests in Africa was published by Sayer et al. (1992). No modeling of deforestation was introduced there although a discussion about the future of African tropical forests was given. Yirdaw (1996) has given a comprehensive review of African deforestation but neither he had any modeling. Palo, Lehto and Enroth (1999) made a pilot modeling of African tropical deforestation. To our best knowledge we have not seen other continental African multiple variable modeling on underlying causes of tropical deforestation so far.

The first purpose of our paper is to identify the role of such underlying factors as population, income, ecological conditions and data reliability on tropical deforestation in Africa.

The second purpose of our paper is to make deforestation scenarios up to 2025 and 2050 in tropical Africa.

The paper is first presenting our empirical data and causal modeling results (Section 2), followed by scenario modeling results (Section 3), and as the last item discussion (Section 4), including conclusions.

2. Empirical data and causal modeling

In this paper we analyze 277 subnational geographical units in 35 tropical African countries. The highest number of units are in Kenya (31), Côte d'Ivoire (26), Tanzania (20), and Nigeria (19), while eleven countries have only one subnational unit and the rest are in between these extremes. We had to exclude Congo, Equatorial Guinea, Guinea, Mali and Niger from our causal modeling due to the lack of parallel data to forest data, but all of them are included in our scenario making.

We used FORIS database of 1995 by FAO (Marzoli 1995) as a source of the variables for analysis. The database has not had a detailed user's manual so far. FORIS consists of both original inventory year by random years from 1970 to 1991, as well as of updated data for the base years 1980 and 1990 of the 1990 assessment by FAO. Only the original forest inventory data were applied in this study. FORIS provides data of forest areas, land areas, ecological zones and human population both at subnational and national levels. Two other parallel data sources were used, namely the United Nations Population Prospects for national populations and NBER Penn World Tables for national income data.

We have been quantitatively modeling pantropical deforestation since 1984. This study has its bases in our second and third generation of modeling (Palo & Mery 1996, Palo, Lehto & Uusivuori 2000, Uusivuori et al. 2002). Our theoretical framework, hypotheses, and expected signs for each independent variable are explained there.

When Palo, Lehto & Uusivuori (2000) experimented with twelve different dependent variables of deforestation and forest cover, and with the same model structure and 16 independent variables as here along with continental dummies, it was easy in this paper to base on those results. FORIS database has no forest change data for Africa, only for Asia and Latin America. Accordingly, we had to cope with forest stock variables. We chose the logistic forest cover as dependent variable in order to simulate best the hypothised descending sigmoid function form (s-shape) of forest cover decrease.

With sixteen variables we were able to explain 46 percent of the variation of the dependent variable or relative forest area (forest area as a percentage of nonforest area). Only four variables did not differ from zero at significance level under ten percent risk. The estimated signs of the significant variables matched the expected plus or minus signs. Population and income were indicated with one variable of the year t and another of the year t-10. While the two variables understandably were strongly mutually correlated, we paid attention for explanation and scenario making to the one, which had a higher absolute regression coefficient (Table 1).

Perhaps the only surprise with these results was that variables 1-2 - the two subnational populations - (Table 1) did not become statistically significant. We assume here that perhaps in Africa the quality of the population statistics has remained too poor because we have a strong hypothesis that subnational population is reflecting the local deforestation effect via subsistence farming and fuelwood gathering. We assume that national populations measure indirect deforestation effect via demand for food and forest products.

The frequency distribution of our dependent variable - forest area as a percentage of nonforest area - is not deviating too much from the normal distribution (Figure 1). The residuals of our modeling of Table 1 are plotted against model estimates (Figure 2). Only a few outlying residuals are deviating outside the length of the whole range of estimates, therefore we consider the variance of the residual distribution is not too wide and, on the other hand, the distribution has no systematic biases. The two qualitative evaluations (Figures 1-2) seem to support the application of our function form and ordinary least squares estimation method (OLS).

3. Scenario modeling

3.1 Trend scenarios

Palo, Lehto and Enroth (1999) made earlier scenario modeling of somewhat similar type with multiple regression but also with trend modeling. Africa was modeled separately along with the two other main tropical continents. Here our regression modeling is more advanced but in the discussion we shall compare the results in both cases. However, we wish to adopt here also the results of the previous trend modeling.

Our trends were of four kinds, namely proportional, linear, accelerating proportional and accelerating linear. In each case we computed the trends as continental decrease and as a sum of subnational units. This gave us eight different trends among which we shall introduce the minimum and maximum values (Figure 3). The computations were based on updated data of 1980 and 1990 by FAO (1993) and 1970 by us.

The trend modeling shows the annual average deforestation continuing at about the present level, 3.4 - 4.4 million hectares from 1990 to 2025, and later on from 2025 to 2050 slightly slowing down at 2.5 - 3.8 million hectares. These trend scenarios assume explicitly no changes in government policies, accessibility, quality of data or in population or economic growth. In practice some allowance is given to changes in these factors because four different function forms were applied.

3.2 Regression scenarios

We applied the regression model of Table 1 to generate seven alternative deforestation scenarios for 40 tropical countries of Africa (Table 2). The five countries excluded from the regression model estimation were included also in this scenario making. The model has sixteen independent variables, of which we changed the values of variables 3 to 8. We applied the low, medium and high national population scenarios by the United Nations (1995) for 1990-2025 and 2025-2050.

Our assumptions on the growth of Gross Domestic Product per capita for the whole period was based on the 1980-1990 annual average -1.1 % (low), which we then raised up in two steps by zero growth alternative (medium), and by + 3.0 (high), which we regarded still feasible. According to our understanding, the future economic growth in Africa is utmost uncertain. Therefore we thought these assumptions could be as realistic as any outcomes from sophisticated modeling.

High reliability of forest data (variable 7 of Table 1) dummy has a high regression coefficient, meaning that it can play some role in the outcome of future deforestation scenarios. Africa has only five countries in reliability class 1 of the state assessments of forest resources by FAO in this dataset. We made an assumption that as an average from 1990 to 2025 a half of the countries would belong to this category, but after 2025 all the countries. The variables 7-9 were excluded from the model from 2025 to 2050.

The assumptions on population and GDP as well as the seven alternative deforestation scenarios are introduced in Table 2. According to these results the tropical forest area in Africa will be reduced by one third to one half until 2025, which would mean an annual average decrease of 4.9 - 8.1 million hectares. This would mean faster deforestation from the present level and also in comparison with our trend scenarios (Figure 3).

From 2025 to 2050 we have assumed that the scale of forest area would become stabilized with no more deforestation. The plausibility of this assumption is based on our belief that Africa would get then most reliable forest resources and other statistics. That would strengthen the countervailing powers in their combat against deforestation at the international, national and local levels. Also better monitoring of changes in forests, population and other relevant data would facilitate more effective forest, land use and other sector policies. Furthermore, absolute population growth is slowing down and if economic growth is taking place a high number of countries will exceed a critical turning point (of so called Kuznets curve), after which economic growth will decelerate deforestation.

4. Discussion

4.1 Comparisons

We consider that our regression model is of relatively good quality (Table 1, Figure 2). Some development possibilities naturally exist. For example, we have found that an increase in agricultural productivity is increasing relative forest area, and increases in openness of external trade and corruption are decreasing relative forest area.

Our present deforestation scenarios can be compared with a few earlier ones. Most interesting is that Palo, Lehto and Enroth (1999) made scenarios with a multiple variable regression model and arrived at an annual average deforestation of 5.7-6.2 whereas our alternative 4, which is the best comparable one, gave 5.3 million hectares from 1990 to 2025. The previous study used a regression model with eight independent variables, such as one national population, one Gross National Product, two ecological zones, the same three reliability variables and national land area. In the previous study we based our estimations on updated forest area data and not on original inventory years as in our present study.

Matsuoka, Morita and Harasawa (1994) applied population density as the only independent variable in making deforestation scenarios. They produced scenarios for 2025 and 2050 as 4.8 and 3.0 million hectares respectively as annual averages. According to their results from 1980 to 2100 some 62 % of the 1980 forest area would be deforested. In our study 32-54 % would be deforested respectively already until 2025 (Table 2). This Japanese team used 1980 forest data, which were of lower quality than our 1990 assessment data.

Gaston et al. (1998) studied the state and change of carbon pools in the tropical forests of Africa at the subnational level based on FORIS database and some other sources. In order to predict anthropogenic effects on biomass they used regressions with population density as the sole independent variable. Their prediction period was limited to 1980-1990.

In this paper we used forest area data of the 1990 assessment by FAO (1993) as complemented by new data until 1995 in the FORIS database (Marzoli 1995). A new Global Forest Resources 2000 Assessment by FAO was completed lately (FAO 2001). However, forest area data from tropical Africa by subnational geographical units and from the original inventory years as reported by the respective countries were not available from this 2000 assessment. Therefore, we used the earlier FORIS data.

4.2 Conclusions

First, in future African studies the most urgent task is to create a more reliable forest resources monitoring system.

Second, the revised forest policy strategy by the World Bank strives to protect environmental services and values provided by forests, harnessing the potential of forests to reduce poverty and integrating forests in sustainable economic development (World Bank 2002). The scenarios of this paper create a serious threat for the implementation of all the three above pillars in Africa.

Third, the Millennium objective of the United Nations aims to reduce the number of poor by 50 % by 2015. Our scenarios will seriously invalidate the use of natural forests in the implementation of this UN Millennium objective.

References

FAO 1993. Forest resources assessment 1990. Tropical countries. FAO Forestry Paper 112. Rome.

FAO 2001. Global forest resources assessment 2000. Main report. FAO Forestry Paper 140. Rome.

Gaston, G., Brown, S., Lorenzini, M. and Singh, K. D. 1998. State and change in carbon pools in the forests of tropical Africa. Global Change Biology 4.

Marzoli, A. 1995. FAO - Forest Resources Assessment 1990. Forest Resources Information System (FORIS). Concepts and methodology for estimating forest state and change using existing information. System documentation. FAO, Rome.

Matsuoka, Y. & Morita, T. & Harasawa, H. 1994. Estimation of carbon dioxide flux from tropical deforestation, Center for Global Environmental Studies, National Institute for Environmental Studies, Tsukuba, Japan.

Palo. M. and Mery, G. 1996. Sustainable forestry challenges for developing countries. Environmental Science and Technology Library Vol. 10, Dordrecht/Boston/London.

Palo, M., Lehto, E. and Enroth, R-R. 1999. Scenarios on tropical deforestation and carbon fluxes. In Palo, M. (ed.): Forest transitions and carbon fluxes. Global scenarios and policies, UNU/WIDER/World Development Studies No 15, Helsinki.

Palo, M., Lehto, E. and Uusivuori, J. 2000. Modeling causes of deforestation with 477 observations. In: Palo, M. and Vanhanen, H. (eds.): World forests from deforestation to transition? Kluwer Academic Publishers/World Forests, Vol. II, Dordrecht/Boston/London.

Sayer, A.J. Harcourt, S. C. and Collins, M. N. (eds.) 1992. The conservation atlas of tropical forests. Africa. IUCN, Cambridge, UK.

Siiriäinen, A. Man and forest in African history. In: Sustainable forestry challenges for developing countries. Eds. M. Palo & G. Mery, Vol. 10 of Environmental Science and Technology Library, pp. 343-357. Kluwer Academic Publishers, Dordrecht/Boston/London, 1996.

United Nations. World population prospects. The 1994 revision. New York, 1995.

Uusivuori, J., Lehto, E. and Palo, M. 2002. Population, income and ecological conditions as determinants of forest area variation in the tropics. Global Environmental Change, Vol. 12, pp. 313-323.

World Bank 2002. World Bank approves new Forest Policy and Strategy. Press Release No: 2003/130/S.

Yirdaw, E. Deforestation in tropical Africa. In: Sustainable forestry challenges for developing countries, eds. M. Palo & G. Mery, Environmental Science and Technology Library, Vol. 10, pp. 291-310, Kluwer Academic Publishers, Dordrecht/Boston/London, 1996.

Table 1: Logistic regression model as estimated from FORIS database (FAO 1993) forest area (% of nonforest area) as dependent variable by 277 subnational units of 35 countries in tropical Africa (all variables in natural logarithms; OLS estimation).

Table 2: Scenario model assumptions and results for tropical Africa up to years 2025 and 2050.

¹ Professor, Finnish Forest Research Institute METLA, Unioninkatu 40 A, FIN-00170 Helsinki, Finland. [email protected]; Website: http://www.metla.fi/wfse